PRISMR: Overcoming Parse Collapse in Multimodal Listwise Ranking via Parameterized Representation Internalization

In three linesPRISMR addresses parse collapse in multimodal listwise ranking with LMMs. Problem: autoregressive decoders silently omit candidates and terminate early on long lists. Solution: lightweight hypernetwork generates item-specific LoRA weights synthesized into instance-specific adapter. New large-scale multimodal review-ranking benchmark introduced.

Summary generated by Claude — human-verified